Neuromorphic Signal Encoding and Decoding using Antiferromagnetic Artificial Neurons

Neuromorphic signal processing is one of the most promising post-Von Neumann computational paradigms [1]. Recently, it was proposed that antiferromagnetic (AFM) spin Hall oscillators, operating in a sub-critical regime, can function as ultra-fast artificial neurons [2, 3]. One of the characteristic features of AFM neurons is the inertial nature of their dynamics: a sufficiently large input spike may induce not one, but several output spikes. Here, we show that this feature can be used for effective encoding and decoding of information from conventional binary to neuromorphic format. The neuromorphic encoder has several inputs, which represent information in parallel binary format, and one output neuron, which produces a train of several spikes. The number of spikes produced by the output neuron equals the input binary code. The decoder reverses this operation, splitting one multi-train input into several binary outputs. The proposed neuromorphic encoder/decoder can be used as an interface between neuromorphic and conventional circuits.

[1] J. Torrejon et al., Nature 547, 428 (2017).
[2] R. Khymyn et al, Sci. Rep. 8, 15727 (2018).
[3] O. Sulymenko et al., J. Appl. Phys. 124, 152115 (2018).